[SOT] Mark dynamic dims by type annotations (#2771)

* [SOT] Mark dynamic dims by type annotations

* fix conflict of forward_meta

* mark more attn backend

* fix missing annotated and add env SOT_SPECIALIZED_DIM_NUMBERS

* auto infer implicit 0 dim dynamic dim

* revert manual marked dims

* revert missing update

* auto infer can use unsafe code in warmup stage

* check -> type_match

* fix codestyle

* restore blank line

* empty commit

* add need_warmup nonlocal;

* add doc for resolver

* add missing type hints

* unquote "ForwardMeta"

fastdeploy/model_executor/layers/attention/append_attn_backend.py

@@ -66,13 +66,13 @@ class AppendAttentionMetadata(AttentionMetadata):
     block_tables: Optional[paddle.Tensor] = None
     rotary_embs: Optional[paddle.Tensor] = None
     attn_mask: Optional[paddle.Tensor] = None
-    encoder_block_shape_q: Optional[paddle.Tensor] = None
-    decoder_block_shape_q: Optional[paddle.Tensor] = None
+    encoder_block_shape_q: int = -1
+    decoder_block_shape_q: int = -1
     _fuse_kernel_compute_dtype: str = "bf16"
 
     # pd_disaggregation
     kv_signal_metadata: Optional[paddle.Tensor] = None
-    kv_signal_data_list: List[paddle.Tensor] = field(default_factory=list)
+    kv_signal_data_list: List[Optional[paddle.Tensor]] = field(default_factory=list)
 
 
 class AppendAttentionBackend(AttentionBackend):
@@ -80,6 +80,9 @@ class AppendAttentionBackend(AttentionBackend):
     AppendAttentionBackend backend implementation.
     """
 
+    __infer_dynamic_dims_fields__ = ["attention_metadata"]
+    attention_metadata: AppendAttentionMetadata
+
     def __init__(
         self,
         fd_config: FDConfig,